index-ai-validator

Appearance

Discovery Checks ​

An agent can only use your AI Manifest if it can find it. These checks inspect the hints that point agents to it.

The validator runs shallow discovery checks through validateIndexAi() and the index-ai CLI.

These checks inspect explicit hints that help agents find the AI Manifest. They do not crawl a site, validate a sitemap, or check DNS TXT records.

Discovery Flow ​

DISCOVERY_HTML_LINK checks the homepage HTML for an explicit agent-manifest link.

Recommended hint:

html
<link rel="agent-manifest" href="/.well-known/index-ai.json" type="application/json">

If the hint is present, the check passes. If it is missing or the homepage cannot be fetched, the check warns.

DISCOVERY_HTTP_LINK_HEADER checks the homepage response headers for an agent-manifest link.

Recommended header:

http
Link: </.well-known/index-ai.json>; rel="agent-manifest"; type="application/json"

If the header is present, the check passes. If it is missing or the homepage cannot be fetched, the check warns.

robots.txt Agent-Manifest Hint ​

DISCOVERY_ROBOTS_AI_INDEX checks /robots.txt for:

txt
Agent-Manifest: /.well-known/index-ai.json

This is a discovery hint. It does not replace crawler rules and does not create legal control over AI agents.

llms.txt Content Type ​

DISCOVERY_LLMS_TXT_CONTENT_TYPE checks that /llms.txt is served as plain text.

Recommended response header:

http
Content-Type: text/plain; charset=utf-8

llms.txt Bridge ​

DISCOVERY_LLMS_TXT_BRIDGE checks that /llms.txt references the AI Manifest.

Recommended text:

txt
- Agent-Manifest: /.well-known/index-ai.json

Mentioning /.well-known/index-ai.json directly also satisfies the bridge check.

Verdict Interaction ​

Discovery checks use SHOULD-level warnings.

  • Present hints pass.
  • Missing hints warn.
  • Discovery warnings do not fail passed by default.
  • failOnWarn makes any warning fail passed.
  • strict makes SHOULD-level warnings fail passed.

Discovery checks do not change structural conformance.

Scope ​

The validator performs shallow, explicit checks only. It does not crawl the site, validate sitemap entries, inspect DNS TXT records, or prove agent adoption. See Scope.